Postal Voting Depends More on the Date Than on the Province, and Requesting It More Has Not Raised Turnout

Requests for postal votes by residents in Spain rose from 2.0% of the electoral roll in 2011 to 7.4% in July 2023. The date of the election explains 53% of the differences between provinces and elections; the province, 36%. It is used mostly by young people, university graduates and residents of big cities, and there is no sign that requesting it more has raised turnout or foreshadows shifts in the vote.

25 October 2026 · Miguel · Leer en español

The window to request a postal vote for the November 29 election opened on October 6 and closes on November 19; ballots must be mailed by November 25 at the latest. There is still no figure for 2026 requests. When one appears, someone will say that a high number heralds higher turnout or a vote leaning one way or the other. With the data from 2008 to 2023 we can say what to expect from such a figure and what not. We are talking about the postal vote of residents in Spain; that of Spaniards living abroad (CERA, a separate system with its own deadlines and its own electoral roll) is a different matter and is left out.

A limit: only requests are published

The postal vote is counted at each polling station at closing time and added to the rest. Official results do not break it out by polling station, municipality or province: it is not known how many postal ballots there were in any place or whom they voted for. What is published is the number of requests by province and election (Oficina del Censo Electoral, the electoral census office of the national statistics institute, INE). In 2023, according to press reports, about 2.47 million ballots were cast before election day, around 94% of the requests. So what follows is about requests, and comparisons between provinces are associations between areas, not between individuals.

A record with a date

Between 2008 and 2015, accepted requests hovered around 2% of the resident electoral roll. After that they rose and fell with the calendar: 4.2% in June 2016, 3.9% in April 2019, 2.9% in November 2019 and 7.4% in July 2023.

The 2,604,386 accepted requests in 2023 are 3.8 times those of November 2011 (687,631), 2.6 times those of November 2019 (997,530) and 1.9 times those of April 2019 (1,351,659). If the ballots cast were about 2.47 million, they amount to almost 10% of that day’s resident voters.

What drives it: the date weighs more than the place

We built a panel of 52 constituencies and 7 elections (364 observations) and decomposed the variance of the logarithm of the request rate per 100 electors.

Relative to 2008, the effect of each election multiplies the rate by 0.87 in 2011 (95% interval: 0.84–0.90), 1.00 in 2015, 1.63 in June 2016 (1.52–1.76), 1.52 in April 2019 (1.39–1.65), 1.20 in November 2019 (1.15–1.25) and 2.58 in July 2023 (2.34–2.84). The two summer elections are the largest, but April 2019 is also above November 2019. With seven elections, the season, the length of the campaign, the 2016 repeat election and a possible habit acquired after the pandemic cannot be told apart: this is a description, not a cause.

Place matters persistently. The correlation across provinces between the logarithms of the rates of two elections ranges from 0.49 to 0.97 (mean 0.74). In the four elections with the lowest use (2008, 2011, 2015 and November 2019) the geographic distribution is almost the same (0.90 to 0.97). The two summer ones, 2016 and 2023, resemble each other (0.87), but 2023 and November 2019 only at 0.55.

Sociodemographic features explain little. With one standard deviation more income per person, the rate is 15% higher (×1.15; 1.05–1.25); with one more in foreign-born population, 11% lower (×0.89; 0.81–0.99). The share of people over 65, the weight of big cities and that of non-primary homes are indistinguishable from zero once the date is accounted for.

Where: big cities and the Basque Country in 2023, the coast at the bottom

In 2023 the rate was 3.1% in Girona and 14.0% in Bizkaia, with a provincial median of 7.2%.

In November 2019 the map was different: Melilla (11.0%) was an exception, followed by small inland provinces (Ceuta 4.7%, Soria 4.5%, Palencia 4.4%, Huesca 4.3%); the lowest were Barcelona (1.8%), Tarragona (1.9%) and Girona (2.1%).

The jump from 2019 to 2023 was not uniform either. For Spain as a whole, requests per 100 electors were multiplied by 2.6; the provincial median was ×2.15, with a range from ×0.9 (Melilla) to ×4.1 (Bizkaia).

The biggest rises were in Bizkaia (×4.1), Madrid (×3.9), Seville (×3.8), Álava (×3.6) and Zaragoza (×3.5); the smallest, in the Balearic Islands (×1.2), Cádiz (×1.4), Almería (×1.4) and Girona (×1.5). Using our own geographic classification (not tourism data), in the 14 provinces of the Mediterranean and south Atlantic coasts and the archipelagos the average rise was ×1.8, against ×2.4 in the other 38; in July 2023 their rate was 4.6%, against 8.3%. In the 13 provinces with the largest population in municipalities of more than 100,000 inhabitants it was ×2.8, against ×2.0 in the 15 with the smallest. Income, foreign-born population and city size together explain 55% of the differences between provinces in the rise.

This is the opposite of what a simple “summer holiday” explanation would suggest: coastal provinces, which receive tourists in July, are the ones that rose the least. We cannot measure who was travelling; one hypothesis compatible with the data is that people who live in a big city and spend the summer elsewhere are the ones who use it most in July. In November that motive disappears.

Who uses it: young people, university graduates, big cities

The CIS (the state polling institute) asked about postal voting in two telephone surveys in 2023: 3411 (June 8 to 27, 29,201 interviews), on requesting it, and 3416 (July 13 to 22, 27,058), on the way of voting. In the second, 13.6% of those who were going to vote had already voted by post (77% of them) or had decided to do so (95% interval: 13.1–14.0; n = 25,069). That is more than actual use, close to 10% of voters, so we use it to compare groups, not to measure the level. For that purpose it works: the CIS percentage by province correlates with the 2023 request rate (r = 0.69; Spearman 0.78; 52 provinces, about 480 interviews in each).

There are three clear differences. Age: 18.3% of those aged 18 to 34 (17.1–19.5) against 9.4% of those aged 65 or over (8.6–10.2). Education: 19.0% of university graduates (18.3–19.7) against 5.0% of those with primary education or less (3.7–6.2). Municipality size: 17.7% in those with more than 400,000 inhabitants against 10.7% in those with up to 2,000. Between men and women there is no difference (13.1% and 14.0%). The June survey, which measures having requested or thinking of requesting the postal vote, gives the same profile: 20.1% among those aged 18 to 34 and 9.1% among those over 65; 19.1% among university graduates and 5.1% with primary education or less.

And the vote: similar, with nuances that depend on age and education

Ideology and 2019 recalled vote show differences, but a good part disappears once age and education are equalized. The table compares the raw percentage with the standardized one: the one each group would have with the same age and education distribution as the whole.

Those who voted for the PP and the PSOE use postal voting in almost equal proportions once age and education are equalized; those who voted for Vox or place themselves at the far-right end of the scale, somewhat less; those who voted for Unidas Podemos and its alliances, somewhat more. The June survey says the same (Vox 10.7% standardized, PP 12.5%, PSOE 14.9%). These are differences of 3 to 7 points on a base of 13%, measured by telephone and with the recall of a vote cast four years earlier.

The CIS also asked which party they had voted for among those who had already voted by post in July. Among those who named one (2,607 interviews), the split was PP 28.0% (26.2–29.8), PSOE 32.7% (30.8–34.6), Sumar (the alliance to the left of the PSOE) 21.7% (20.0–23.3), Vox 8.3% (7.2–9.4) and others 9.4%. The intention of those who were going to vote in person (18,882 interviews): PP 30.0% (29.3–30.7), PSOE 34.0% (33.3–34.7), Sumar 16.3% (15.7–16.8), Vox 11.7% (11.2–12.1) and others 8.1%. The largest differences are Sumar (+5.4 points among postal voters) and Vox (−3.4). Part of this reflects the age and education mix, and we are comparing a vote cast with an intention.

By province, the request rate does not foreshadow shifts in the vote

With the 52 constituencies we looked at the relationship between the request rate and the vote for each party. It is an association between areas and says nothing about what each voter does.

• Between provinces, in each election: the correlation with the PP ranges from 0.04 (2023) to 0.65 (2015); with the PSOE, from −0.22 (2023) to 0.24 (2015); with Vox, from −0.19 (2023) to 0.17 (November 2019). If the five sociodemographic features are controlled for, the PP comes out higher where there are more requests in six of the seven elections (from 3.8 to 8.2 points per standard deviation), but the association depends on what is controlled for.
• Within each province, between elections, there is no signal. One more point of rate is associated with −0.06 points of vote for the PP (−0.60 to 0.48), −0.21 for the PSOE (−0.46 to 0.05) and −0.06 for Vox (−0.35 to 0.22). With changes between consecutive elections, the same: −0.41 (−1.06 to 0.23), −0.20 (−0.51 to 0.12) and 0.08 (−0.27 to 0.43).

More requests, more turnout?

If requesting the postal vote encouraged people to vote who would otherwise not, each extra request would be roughly one extra voter: a slope close to 1 between the change in requests and the change in turnout. If it merely replaced voting in person, it would be close to 0. We measure the turnout of residents (BOE voters minus CERA voters, over the CER electoral roll, that is, the resident electorate).

Within a single election, the correlation between provinces changes sign: from −0.37 (November 2019) to +0.28 (2016); only those two extremes are distinguishable from zero. In 2023, +0.15. More informative is to look at how both change between consecutive elections, province by province.

Only two of the six pairs are distinguishable from zero, with opposite signs. Pooling the six (312 observations) and removing the mean change of each pair, the slope is 0.00 (−0.47 to 0.48; R² = 0.00); with province and election fixed effects, 0.13 (−0.28 to 0.53). A slope of 1 is ruled out. At the national scale, with six changes, it is 0.26 (−2.6 to 3.1; R² = 0.02), and in levels with all seven elections, −0.47 (−1.9 to 0.95; R² = 0.13), not very informative with such a small sample.

The clearest case is 2023: requests rose 4.6 points from November 2019 and the turnout of residents, 0.5 (from 69.9% to 70.4%). Between April and November 2019 turnout fell 5.9 points with only one point fewer requests. Turnout moves for other reasons; in these data, requests neither add to it nor subtract from it appreciably.

What a high figure on November 29 would and would not say

• It would not herald higher turnout. In no analysis (between provinces, between consecutive elections or at the national scale) is an increase in requests systematically accompanied by an increase in turnout.
• It would not herald a result. Between provinces, those with more requests vote somewhat more for the PP once their sociodemographic features are accounted for, but that is a stable trait of those provinces: within each province, whether requests rise or fall is not associated with changes in the vote. Postal voters resemble in-person voters in their preferences, with differences of a few points (Sumar somewhat more, Vox somewhat less) that partly reflect age and education.
• It would not be comparable with July 2023. November 29 is more like November 2011 (2.0%), December 2015 (2.3%) and November 2019 (2.9%) than the summer of 2023 (7.4%). If the date remains what weighs most, a figure closer to those three than to the 2023 one is what to expect, although the underlying rate has been rising (3.9% in April 2019) and seven elections do not allow us to know by how much.

It would say something about the polling stations on the 29th: how many postal ballots will arrive and be counted at closing time. If the geographic distribution resembles that of the lower-use elections (correlations of 0.90 to 0.97 among them), one can expect a map like that of 2019: higher rates in small inland provinces and lower in Catalonia and on the coast. It is an expectation based on four elections, not a forecast. And since there is no official count of postal ballots cast, it will not be possible to check against the results by polling station.

How we calculated it

• Requests. Submitted and accepted by constituency and election, 2008–2023, from the Oficina del Censo Electoral (INE), over the resident electoral roll (CER, excluding CERA); oce.py. We use the accepted ones: the submitted ones are only 0.4% to 1.1% more. The national series is the sum of the 52 provinces. The 2,622,808 admitted requests and the 2.47 million ballots are press figures from July 2023 (a note from Correos, the state postal service, of July 14 for the former) that we could not check by province.
• Turnout of residents. BOE voters minus CERA voters (data/cera/), over the CER electoral roll.
• Panel. 52 × 7 observations; election and province fixed effects (in a balanced panel, the parts add up to the explained variance). Covariates by constituency standardized within each election: % over 65, net income per person (logarithm), % foreign-born, % of population in municipalities of more than 100,000 inhabitants and % of non-primary homes. Errors clustered by province.
• Changes and parties. Slopes with 95% CIs; parties grouped by families (familias.py); Vox only from April 2019.
• CIS. Microdata from studies 3411 and 3416, weighted with the CIS weighting variable; 95% CIs with the Kish effective sample size. Standardization by age and education is direct, with 16 groups (4 × 4), and its CIs come from 200 resamples.
• Limits. Requests, not ballots cast; associations between areas, not between individuals; seven elections are not enough to separate date, season and trend; the CIS sample is telephone-based and overstates use.
• Code. voto_correo.py and articulo_correo.py, in the altnalysis repository.